Put governed AI work in operators' hands with Firebase

Firebase recommends Cloud Firestore as the starting point for new applications that need richer document models, indexed queries, automatic scaling, and offline support across Apple, Android, and web clients. Realtime Database remains useful for simple JSON-shaped state, very low-latency synchronization, or native presence signals. MetaCTO usually anchors a review queue in Firestore, uses Realtime Database only for a specific presence or synchronization need, and keeps the authoritative workflow instance and business record outside both.

No. Firestore mobile and web requests can be authorized and validated with Security Rules, but Firebase documents that server client libraries bypass those rules and authenticate through Google Application Default Credentials; server access must be constrained with IAM. MetaCTO treats the operator's client permission and the service account's write authority as separate boundaries, then gives the write-back service only the destination actions its workflow is allowed to perform.

Firestore transactions are atomic and can rerun after a concurrent edit, but transaction functions may execute more than once and transactions fail while the client is offline. With offline persistence, cached reads can be stale or incomplete and multiple changes to the same document resolve last-write-wins when synchronized. MetaCTO therefore records a source version and approval precondition, rechecks current state on the server, and routes stale or conflicting decisions to review instead of trusting the latest screen state.

Firebase states that Firestore trigger ordering is not guaranteed and events are delivered at least once, so one event can invoke a function more than once. The safe boundary is an idempotent handler keyed by the work item and intended operation, with a durable receipt from the destination system and a terminal success or exception state. MetaCTO does not treat a function invocation or a changed Firebase status as proof that the business action completed.

Firestore supports scheduled backups and point-in-time recovery for accidental deletion or modification, but recovery is still an operating procedure rather than an automatic reconciliation of downstream systems. MetaCTO identifies which queue, approval, and evidence records are recovery-critical, tests restoration, and then reconciles restored Firebase state against the durable workflow and each system of record before operators resume action. Notifications remain attention signals; persisted queue state determines what is actually open or overdue.